Quote:Taiwan Semiconductor Manufacturing Company ("TSMC") has quickly become a household name in the chip fabrication business, and the foundry of choice for many fabless chip makers. In recent times, TSMC has eclipsed Intel when it comes to advancing research and investing into smaller, more efficient process nodes, and at the volumes necessary to support global demand. While Chipzilla is still struggling with its 10nm process, TSMC has already hit volume production for parts using its 7nm process and is expecting over 50 tape-outs this year alone.
...
At the conference, TSMC also announced a new 7nm+ node using extreme ultraviolet (“EUV”) lithography, which should ramp up during H1-2019, and that it’s on track to start ‘risk-production’ of a 5nm node, also during H1-2019 for mobile and high-performance computing chips. Back in January, TSMC started building a brand new 5nm fab in Taiwan, which is expected to start volume production in 2020. There's a lot to look forward to in H2-2018, particularly the next-gen 7nm mobile SoCs, CPUs and GPUs, and what they have in store in terms of performance and power efficiency.

Quote:At its 24th annual Technology Symposium in Santa Clara, TSMC announced its Wafer-on-Wafer (“WoW”) technology, which could be a boon for GPU performance, similar to how vertical stacking has improved DRAM manufacturing and performance. The technology, as the name suggests, stacks layers of logic on top of one another using a through-silicon via (“TSV”) interconnect. With limited lateral space on wafers, WoW technology should allow for faster chips with higher areal density to be crammed in the same amount of space using high-speed, low-latency interconnects.

Process maturity plays an important role in determining yields, which are currently low for the WoW tech – so expect an initial roll-out on parts built on more mature 16nm or 10nm processes, before transitioning to smaller process nodes. However, as the technology matures and yields improve, GPU manufacturers could literally stack two fully-functional GPUs on top of one another, as opposed to dual-GPU setups using two independent dies, leading to cost savings, and smaller, more power-efficient video cards.

Quote:We’ve discussed AMD’s 7nm plans several times this year, including the company’s decision to source its 7nm silicon for Vega’s machine learning iteration from TSMC. As Extreme Ultraviolet Lithography (EUV) ramps up and begins to enter production, a number of foundries including TSMC, Samsung, GlobalFoundries, and Intel are pursuing their own strategies and introduction plans. One interesting tidbit in a recent report sheds a bit more light on AMD’s 7nm plans and suggests we’ll see the company pursuing dual-source strategy well into next year.

In an article at EETimes, GlobalFoundries admits using similar pitches and SRAM cells to TSMC’s 7nm, precisely because it gives AMD more flexibility. The company’s chief technologist, Gary Patton, told EET’s Rick Merritt that AMD “will have more demand than we have capacity, so I have no issues with that.” The same article also confirmed that AMD will provide its first 7nm chip with tapeout expected later this year. IBM hardware and other ASICs will follow in 2019.

Quote:Digitimes reports that TSMC CEO CC Wei told onlookers that commercial production of chips built using the company's 7-nm fabrication process has begun at a recent technology symposium. The leader of the Taiwanese foundry says an improved 7-nm node with EUV will come before the end of 2018, and it anticipates a move to a 5-nm fabrication node at the end of 2019 or in early 2020. The CEO said TSMC's newest plant, Fab 18 in Taiwan Science Park, will also be the tip of the spear for 3-nm production at some undetermined point in the future, according to Digitimes.

Quote:AMD also announced that it is shifting all of its 7nm products to TSMC foundries, while long-time partner GlobalFoundries has announced that it is terminating work on the 7nm process.

Quote:TSMC, the world's biggest contract semiconductor manufacturer, who is at the forefront of 7 nanometer production has just announced that they are making good progress with their second generation of 7 nm technology "N7+", using EUV (Extreme Ultraviolet Lithography). A first design for N7+ from an unnamed customer has been taped out. The company's first-gen 7 nm production is running well already, with final products, like Apple iPhone already in the hands of customers.

While not fully EUV yet, the N7+ process will see limited EUV usage, for up to four non-critical layers, which gives the company an opportunity to figure out how to make best use of the new technology, how to ramp up for mass production and how to fix the little quirks that show up as soon as you move from the lab to the factory.
...
Moving beyond 7 nanometers, TSMC's target is 5 nm, internally called "N5". This process will use EUV for up to 14 layers and is expected to be ready for risk production in April 2019. According to TSMC, many of their IP blocks are ready for N5, with the exception of PCIe Gen 4 and USB 3.1. We have all been waiting for PCIe Gen 4 on new GPUs and it looks like we'll have to wait even longer for this new version to become available. Compared to the N7 designs, which have initial costs in the $150 million range, the cost for N5 is expected to increase even further, up to $200 to $250 million.

Quote:By the end of 2018, TSMC will have taped out 50 7 nm designs, and plans to double that number in 2019. And these design wins don't stand solely on the shoulders of TSMC's first 7 nm technology (which should account for 20% of the company's revenue by 2019); the company will also tape-out chips built upon their 7 nm + EUV process, which will begin production in 2019.

Quote:Due to a cutback in orders placed by Apple, HiSilicon and Qualcomm, concerns regarding TSMC's ability to be the sole 7 nm chip fabrication tech for the industry can likely be laid to rest. That the smartphone market is reaching saturation is a well-known quantity - it's becoming harder and harder to cram new technologies that justify the yearly smartphone upgrade that most companies vie for - and one of the reasons for the launch of various brand-specific smartphone subscription services. The difference isn't scandalous - TSMC will still be making use of 80-90% of its total 7nm process capacity during the first half of 2019, the report quoted industry sources as saying.

Quote:TSMC is giving final touches to set its flagship 7 nanometer EUV (extreme ultraviolet lithography) silicon fabrication node at its highest state of readiness for business, called mass-production. At this state, the node can mass-produce products for TSMC's customers. TSMC had taped out its first 7 nm EUV chips in October 2018. The company will also begin risk-production of the more advanced 5 nm node in April, staying on schedule. Mass production of 5 nm chips could commence in the first half of 2020.

Quote:TSMC hit a rough patch a few months back. The company announced in August 2018 that it could face production delays after a variant of WannaCry infected some of its servers. Then, in January it revealed that a chemical contamination at Fab 14B resulted in at least 10,000 defective wafers. TSMC had to cease production, investigate the extent of the issue and ultimately revise its earnings guidance in response.

But today's DigiTimes report indicated that demand for 7nm chips might help TSMC recover from those setbacks. The report claimed that, between HiSilicon and AMD "ramping up their wafer starts aggressively" and rising demand from Android smartphone makers, TSMC expects to see its 7nm manufacturing processes run at full capacity in the third quarter of 2019. All of this according to anonymous industry sources, of course.

Quote:TSMC’s CEO said the company has started volume production of its 7N+ process technology, DigitTimes reported on Friday. It is the company’s first process node to use EUV on a few critical layers. The company says the yield is on par with 7nm.

In February, DigitTimes had reported that volume production would start in late March, so it appears that it has been delayed by up to two months. The company did not mention what the first 7N+ product will be, but it is rumored that Huawei’s Kirin 985 will be built on 7N+. The node will provide a 20% increase in transistor density and offers 10% more performance or 15% lower power than the first generation 7nm.

Quote:The Taiwan Semiconductor Manufacturing Company (TSMC) has silently introduced new performance-enhanced versions of its 7nm DUV (N7) and 5nm EUV (N5) manufacturing processes, called N7P and N5P, respectively.

TSMC’s 7NP performance-enhanced process, which is DUV-based, is not to be confused with the company’s N7+, which is TSMC’s first process to use extreme ultraviolet lithography (EUV). The N7P is simply an optimization of N7 process; it uses the same design rules and is fully IP compatible with N7.
...
As with the N7P, TSMC plans to offer a performance-optimized version of N5, its next-generation “full node,” which is expected to be used by most of TSMC’s customers in the near future once they switch from the 7nm processes. The process uses the same design rules and if is fully IP compatible with the N5 EUV process.

Quote:It looks like TSMC is the belle of the ball, and everyone wants a dance. DigiTimes today reported that, according to anonymous industry sources, the semiconductor manufacturer's lead time for 7nm chips has tripled from two months to six months because of high demand.

Right now AMD is the most well-known purchaser of 7nm chips from TSMC--the company relies on those chips in Ryzen 3000 processors and Navi graphics. Nvidia told us in July that it will make its next-gen GPUs with both TSMC and Samsung, meaning the two largest graphics companies will be vying for the manufacturer's 7nm chips, assuming the reports about Nvidia adopting a 7nm process are true.

Quote:AMD has its entire modern product stack built under the 7 nm process, so this could potentially affect both CPUs and GPUs from the company - and let's not forget AMD's Zen 3 and next-gen RDNA GPUs which are all being designed for the 7 nm+ process node. TSMC is expected to set aside further budget to expand capacity of its most advanced nodes, whilst accelerating investment on their N7+, N6, N5, and N3 nodes.

Quote:According to industry sources over at DigiTimes, TSMC will begin mass production of its 5 nm node in March 2020, when companies using the 5 nm PDK can tape out their designs and integrate them into future products. Going into volume production two years after the 7 nm node, 5 nm is trying to put Moore's Law back on track again.

Quote:Semiconductor market research firm IC Insights predicted yesterday that TSMC's sales would jump 32% year-over-year in the second half of 2019. The firm expects the rest of the integrated circuit market's sales to rise by 10% in that same period, meaning that TSMC'could experience growth at over three times the rate of the rest of the market. Much of that extra growth can be attributed to the popularity of processors made with its 7nm process.

That popularity was no secret: DigiTimes reported earlier this month that the lead time for TSMC's 7nm chips tripled from 2 months to 6 months because of high demand. The publication followed up that report today with a new one claiming TSMC had told its clients to book their 7nm capacity for the entirety of 2020. We wouldn't be surprised if companies are forced to fight over increasingly limited supplies.
...
But this could be nothing compared to what happens when TSMC introduces the 5nm process. IC Insights said that companies have been quicker to adopt new process technologies: "It took eight quarters for the foundry’s 40-45nm technology to secure greater than 20% of its total sales, five quarters for its 28nm process to exceed that threshold and only three quarters for its 7nm process to account for more than 20% of its quarterly revenue."

Quote:TSMC and Arm announced they developed the industry’s first 7nm chiplet system. The proof-of-concept HPC system is based on multiple high-speed Arm cores and a mesh bus, using TSMC's new LIPINCON inter-chiplet interconnect with CoWoS packaging technology.

Chiplet-based designs are one of the current industry trends. This could be connecting various chiplets to a base die, but for high-performance computing (HPC) it often involves partitioning a big die into multiple smaller chiplets, then using advanced packaging technology to achieve performance as close as possible to monolithic integration. The benefit is improved overall yield, and thus cost, due to the smaller silicon dies, although the advanced packaging cost can offset the savings somewhat.
...
TSMC says the system taped out in December 2018 and was produced in April 2019 (which gives a rough indication of 7nm's cycle time). TSMC provided more details of the design at the 2019 VLSI.

Quote:Graphics card and CPU vendors may be left fighting over 7nm scraps. Nikkei Asian Review reported today that Apple told suppliers to increase production of the iPhone 11, iPhone 11 Pro and iPhone 11 Pro Max by 10%, due to higher-than-expected demand. That comes out to roughly 8 million additional units--all of which rely on the custom A13 Bionic chip manufactured using TSMC's already-strained 7nm process.
...
That could pose a problem for TSMC's other customers as they pursue their own 7nm ambitions. AMD uses the manufacturer's chips in its Ryzen 3000 CPUs and Navi graphics cards. Nvidia has said that it will source its first 7nm graphics cards from both TSMC and Samsung. Enthusiasts might not typically care much about the new iPhone, but that might change if next-gen graphics or CPU supplies are impacted by Apple's latest.

Quote:TSMC today announced that its N7+ process is going to market in high volume, and the company already has customers lined up, including AMD, which has stated publicly that it will use the new 7nm+ process for its future Zen 3-based chips.

It’s not "just" TSMC’s first process that uses the long-awaited extreme ultraviolet (EUV) lithography, TSMC also claims that N7+ is the industry’s first commercially available EUV process. The company did not clarify which products will use the new chips.
...
TSMC claims that its N7+ production is one of its fastest ever, which presumably means in terms of yield learning, as TSMC then goes on to say that N7+ is already matching yields of the original 7nm process from 2018 that debuted with Apple’s A12 chip. N7+ went into volume production in May and will be followed by the slightly improved N6 by the end of 2020, with similar density of N7+ but full design rule compatibility to N7.

Quote:Building on its successful experience, N7+ sets a path for future advanced process technologies. TSMC will bring N6 technology into risk production in the first quarter of 2020 for volume production by the end of the year. With further application of EUV, N6 will offer 18% higher logic density over N7, and design rules fully compatible with N7 enable customers to greatly shorten time-to-market.

Quote:It’s good times to be in the foundry business — at least if you are TSMC. The firm has announced plans for much higher capital expenditures than it originally forecast for this year. Originally, TSMC expected to spend ~$11B on CapEx improvements for 2019. Now, the company will spend between $14-$15B after allocating another $4B in cash to increased 7nm and 5nm production.

7nm accounted for 27 percent of TSMC’s quarterly revenue, with 10nm at 2 percent and 16nm at 22 percent. Nodes at or below 16nm accounted for 51 percent of the company’s revenue, while older nodes contributed 49 percent. According to TSMC, the increased production need on 7nm and 5nm is being driven principally by 5G deployment. Capex in 2020 is expected to be in the $14-15B dollar range as well.
...
6nm is a follow-on to 7nm+. Design layout rules are said to be fully compatible with 7nm+, with a further 18 percent improvement to density. No word on any other improvements, but this node uses one more EUV layer than 7nm+. Risk production in Q1 2020, volume production before the end of the year. No customer announcements for this node yet.

Quote:A DigiTimes premium report, interpreted by Chiakokhua, aka Retired Engineer, chronicling NVIDIA's move to contract TSMC for 7 nm and 5 nm EUV nodes for GPU manufacturing, made a startling revelation about NVIDIA's recent foundry diversification moves. Back in July 2019, a leading Korean publication confirmed NVIDIA's decision to contract Samsung for its next-generation GPU manufacturing. This was a week before AMD announced its first new-generation 7 nm products built for the TSMC N7 node, "Navi" and "Zen 2." The DigiTimes report reveals that NVIDIA underestimated the efficiency gains AMD would yield from TSMC N7.

With NVIDIA's bonhomie with Samsung underway, and Apple transitioning to TSMC N5, AMD moved in to quickly grab 7 nm-class foundry allocation and gained prominence with the Taiwanese foundry. The report also calls out a possible strategic error on NVIDIA's part. Upon realizing the efficiency gains AMD managed, NVIDIA decided to bet on TSMC again (apparently without withdrawing from its partnership with Samsung), only to find that AMD had secured a big chunk of its nodal allocation needed to support its growth in the x86 processor and discrete GPU markets. NVIDIA has hence decided to leapfrog AMD by adapting its next-generation graphics architectures to TSMC's EUV nodes, namely the N7+ and N5. The report also speaks of NVIDIA using its Samsung foundry allocation as a bargaining chip in price negotiations with TSMC, but with limited success as TSMC established its 7 nm-class industry leadership. As it stands now, NVIDIA may manufacture its 7 nm-class and 5 nm-class GPUs on both TSMC and Samsung.

Quote:TSMC is celebrating a recent silicon milestone — one billion 7nm chips manufactured. That’s good die, not total manufacturing volume, and it’s enough to cover more than 13 Manhattan blocks. TSMC’s 7nm foundries are, according to TSMC, running beautifully.
...
TSMC has effectively taken a leadership position in the entire foundry industry. With Intel’s 7nm having slipped out into 2022, and Samsung apparently having trouble matching them in terms of overall efficiency, the Taiwanese manufacturer is putting its own imprimatur on the leading edge.

Quote:It's no secret that AMD's Zen-3 based Ryzen 5000 series CPUs are flying off the shelves faster than AMD can make them, but it looks like the same will be the case for Zen 3 notebook chips when they land in Q1. A report from DigiTimes doesn't specifically call out the reason behind the issues, but claims they could be due to either PlayStation 5 orders or a lack of packaging and substrate availability.

But the latter issue doesn't only affect AMD. Reports show that Nvidia, Apple, Huawei, EV cars, and virtually any other market that uses silicon interposers are suffering at the hands of the shortage of critical chip packaging supplies.
...
Naturally, in such a situation, AMD is quite powerless as a fabless chipmaker because it relies on TSMC for its parts. One could argue that TSMC's suppliers should increase supply, but that would require increased capacity and investments from the substrate makers. No one can predict with any certainty that post-pandemic demand will remain at this level, potentially making that a poor investment for those firms, so that doesn't seem likely.

This means the current difficulties with finding Ryzen 5000, Radeon RX 6000, PS5, and Xbox Series X/S consoles are likely to continue. And the same goes for Nvidia's RTX 30-series GPUs.

Quote:ABF is incredibly important to the chip-packaging process, and an ongoing shortage is hitting pretty much everyone who employs advanced packaging standards. This could fit our criteria for an important piece of the overall explanation of what’s going on, because this isn’t an issue that would impact just Nvidia, AMD, Intel, or any other single company. If TSMC can’t buy enough of it, the impact could ripple out across the market, hitting a number of companies. Apple, Qualcomm, and Samsung all use ABF as well. DigiTimes reported on a shortage at least as early as June 2020 and claimed it could widen and worsen in 2021. That prediction seems to have borne fruit.

Low yields on GDDR6 have also been blamed for GPU shortages, though again, it’s not clear exactly which component shortages are driving which problems. But it’s clear this goes beyond yield problems, which makes sense — we have heard rumors of low 8N yields at Samsung, but nothing about any equivalents at TSMC. We also know that AMD has been shipping 7nm silicon out of TSMC since mid-2019, implying that the node should be pretty mature by now, even for high-power desktop chips and GPUs. The fact that we’re seeing global problems does suggest there might be additional variables in play.

Also, the crypto market is exploding again, which is going to help exactly no one and nothing when it comes to finding a solution to ongoing shortages.
...
TSMC’s major suppliers for ABF are all rumored to be experiencing ongoing shortages. There are rumors that AMD can’t currently meet demand for notebooks because of the ABF problem and that the issue could worsen in Q3 2021 when Zen 3 notebooks come to market. Up until now, we’ve heard a lot of rumors that the shortages would ease in March-April 2021, but if the ABF angle is true, it could take longer.